Jupiter: How a Non-Programmer Built Windows Dictation with ChatGPT

⚡ The gist in 5 seconds - An actor and writer with no programming skills created his own voice dictation tool, Jupiter, for Windows using ChatGPT. - The project runs locally: your voice never goes to the cloud, and recognition happens on your computer. - The author admits it's a personal experiment: he doesn't plan to release it as a product and believes nobody needs it. ### 🔍 What was found On September 30, a user named Trop_12 posted a case study on Habr about how he, despite not being a programmer, built his own voice dictation app. The author is an actor, model, writer, financier, and entrepreneur. Tired of typing so much, he tried Windows' built-in Voice Typing (Win+H). The result didn't satisfy him: it was slow, laggy, entered words with a delay, and hogged the CPU. Other programs were also inconvenient, and most of them send dictated text to the cloud — unacceptable for the author. So he decided to build his own tool. Jupiter came to life through a dialogue with ChatGPT: the author ran the code, got errors, sent them to the neural network, pasted in the fixes — and so on in a loop. The system listens to the microphone, a separate layer detects when a person starts speaking, recognizes the stream, and inserts text wherever the cursor is. It works in any application — Word, Telegram, a browser, Notepad. Recognition is streaming, so the reconciliation layer corrects already-typed text if the model refines its recognition. The author deliberately avoided using one big neural network: instead, he assembled the system from small specialized components. This lets Jupiter run even on a weak Windows PC. ### 💡 Why it matters This case shows that modern AI tools allow people without technical education to build working solutions. The author doesn't know algorithms, design patterns, or Clean Architecture, but thanks to ChatGPT he was able to take the project to the point where it stopped being just an experiment. His architectural choices, such as simulating keyboard input and local processing without the cloud, could be useful for privacy-sensitive scenarios. Moreover, Jupiter disproves the notion that dictation requires a huge neural network — a set of small models is enough. ### 🧩 Context The author writes a lot and gets tired of typing. After trying standard tools, he was unhappy with their speed, quality, and privacy. So he decided to build his own dictation tool. After months of experimentation, Jupiter began working reliably. Still, the author doesn't consider it a market achievement: he simply enjoys having brought his idea to life. But the very fact that such a project was within reach of a non-specialist is a vivid illustration of what AI makes possible.

🔗 Read on habr.com